split window algorithm

Soil temperature is a key factor that controls physical, chemical and biological properties of soil and its processes. Since soil temperature is measured at synoptic stations and data availability, especially in arid lands, is limited, capability of satellite images to estimate soil temperature at different depths evaluated in the Yazd-Ardakan basin, as the study area. Daily soil temperature at 5, 10, 20, 30, 50 and 100 cm depth measured at synoptic stations of Yazd, Meybod, and Mehriz for the periods of 2014 to 2016, and Landsat 8 satellite images of were used as the main data in this research. Then, using split-window surface temperature, Land Surface Temperature (LST) maps were estimated. Temperature trend from soil surface to a depth of 100 cm were examined seasonally. Using simple linear regression and artificial neural network techniques, the relationship between temperature of surface soil and soil temperatures at different depths were predicted. Results showed that the artificial neural networks had greater accuracy than the linear regression method in all seasons. The lowest accuracy of this method is related to the soil temperature at 5 cm depth. Artificial neural networks can be used for predicting of soil temperature till depth of 100 cm, using land surface temperature obtained by Landsat 8 images. To validate the results, soil temperatures at depth of 30 cm for 16 selected points in the study area were compared with estimated soil temperature using Landsat images and artificial neural network. Absolute error of measurements show that the maximum error was observed to depth of 30 cm (3.7 ℃). Therefore, using the measured soil surface temperature by applying the split-windows and artificial neural network can be used to predict soil temperature.